Is a confidence interval the same as a prediction interval?

No; a prediction interval concerns an outcome to be predicted rather than only a parameter.

AI Glossary / Statistics and probability / Advanced

Interval estimate

Definition

An interval estimate is a range used to express uncertainty about an unknown parameter, such as a population mean or a difference between groups. Its lower and upper limits come from an estimation procedure. The range needs an identified target, method, and interpretation; two numbers alone do not explain what uncertainty has been quantified.

What it means

A point estimate gives a single value. An interval estimate adds information about how precisely that target has been estimated under specified assumptions.
The interval is about the target parameter, not automatically about individual observations. An interval for an average task duration does not promise that most individual tasks finish within those limits. The target must stay visible whenever an interval is reported.

How it works

Identify the quantity being estimated and choose a procedure suited to the sampling design and statistical model. A common confidence-interval construction combines a point estimate, its standard error, and a critical multiplier. Other methods can produce asymmetric limits or use resampling.
For an independent normal sample with unknown population standard deviation, the usual interval for the mean uses the sample standard deviation divided by the square root of sample size and a Student t critical value with n minus one degrees of freedom.
A frequentist confidence level describes the procedure's repeated-sampling coverage under its assumptions. A Bayesian credible interval instead describes posterior probability conditional on a model, prior, and data. Their interpretations differ even when endpoints agree.

Three practical examples

A service team estimates average task duration and reports an interval for that mean. It separately examines the spread of individual task durations.
A randomized experiment reports an interval for the difference in average outcomes between assigned groups. The interval helps assess both direction and decision-relevant magnitude.
A model-evaluation team estimates an accuracy difference with uncertainty appropriate to its evaluation design. Repeated predictions on the same cases are not automatically independent observations.

Think of it this way

A point estimate marks one location on a ruler. An interval adds limits produced by a particular uncertainty procedure. To understand those limits, you need to know what is being located and how the ruler's uncertainty was assessed.

Worked example

Consider an invented measurement study with sixty-four independent observations from a normal population. Suppose the observed sample mean is 12 units and the population standard deviation is known to be 4 units. The known-standard-deviation assumption is stipulated for this teaching example; it is not inferred from the sample.
The standard error of the mean is 4 divided by the square root of 64 = 0.5 units. A two-sided normal 95 percent confidence interval uses approximately 1.96 standard errors on each side. The margin is 1.96 times 0.5 = 0.98, giving limits 11.02 and 12.98.
The full interval width is 12.98 minus 11.02 = 1.96 units. Its center is the sample mean, 12. These limits describe uncertainty about the population mean under the stipulated sampling model.
Why does the procedure have approximately 95 percent coverage? If the true mean is mu, the standardized quantity (sample mean minus mu)/0.5 has a standard normal distribution. The event that mu lies between sample mean minus 0.98 and sample mean plus 0.98 is equivalent to this standardized quantity lying between minus 1.96 and 1.96. That event has probability approximately 0.95 before sampling.
After this particular interval has been calculated, the fixed population mean either lies inside it or does not. The frequentist calculation has not assigned a posterior probability distribution to that fixed mean.
Now request 99 percent confidence using an approximate multiplier of 2.576. The margin becomes 2.576 times 0.5 = 1.288, and the interval becomes 10.712 to 13.288. Greater coverage requires wider limits here; it does not produce a more precise estimate from the same observations.
Next imagine a separate study with 256 independent observations, the same known standard deviation of 4, and, for comparison, the same observed mean of 12. Its standard error is 4/16 = 0.25. The 95 percent margin is 0.49, giving 11.51 to 12.49. Quadrupling the sample size halves the width in this particular setting.
Finally, suppose the original measurement device adds a systematic error of 2 units to every reading. The interval 11.02 to 12.98 then concerns the mean recorded value under the model as applied. If the offset is known exactly and corrected, the mean estimate becomes 10 and the interval becomes 9.02 to 10.98. Increasing sample size alone would not remove that offset. If the correction itself is uncertain, its uncertainty must also be addressed.
The example therefore separates three issues: sampling precision, the chosen coverage level, and systematic measurement error. Changing one does not automatically solve the others.

Illustrative case study

Fictional company Maple Evaluation compares two assistant workflows. Its report identifies the outcome, the population being studied, the estimated difference, and the interval method. Managers review whether the limits include differences large enough to affect staffing decisions.
When an interval is wide, the team avoids treating the point estimate as a precise operational forecast. It also checks whether repeated tasks from the same employees require an analysis that accounts for dependence. This scenario is fictional.

Common misconceptions and mistakes

A 95 percent confidence interval is not a statement that 95 percent of the individual measurements lie inside it.
A narrow interval can still accompany a biased measurement process or an unsuitable sample.
An interval containing zero does not prove that the true difference is zero.
The confidence level does not automatically cover uncertainties omitted from the model, such as an unexamined measurement change.
Intervals from different procedures should not be compared without checking their targets, assumptions, and coverage or probability interpretations.

Business use

Report intervals with the point estimate, unit, target population, method, and relevant assumptions. Interpret the limits against practical thresholds rather than reducing the result to whether zero is included. Where bias, dependence, or missing data matter, explain how the analysis addresses them and what uncertainty remains outside the reported limits.

Check your understanding

What is the original standard error?

Answer: 0.5 units.

What happens to the width when the stipulated sample size rises from 64 to 256?

Answer: It halves.

Does the uncorrected interval remove a systematic two-unit offset?

Answer: No.

Related terms

Learn next

Study interval construction for proportions and differences, dependence-aware uncertainty, and the distinction between sampling uncertainty and systematic error.

Sources

Howell, Statistical Methods for Psychology glossary: https://www.uvm.edu/~statdhtx/methods8/Glossary/Glossary.html . NIST, Confidence Limits for the Mean: https://www.itl.nist.gov/div898/handbook/eda/section3/eda352.htm . Numerical study and company scenario are original illustrations.

Previous: Interval algebra Next: Interval predictor model